Essential Concepts in Toxicogenomics

Essential Concepts in Toxicogenomics

M E T H O D S

I N

M O L E C U L A R

B I O L O G YTM

John M. Walker, SERIES EDITOR 470. Host-Pathogen Interactions: Methods and Protocols, edited by Steffen Rupp and Kai Sohn, 2008 469. Wnt Signaling, Volume 2: Pathway Models, edited by Elizabeth Vincan, 2008 468. Wnt Signaling, Volume 1: Pathway Methods and Mammalian Models, edited by Elizabeth Vincan, 2008 467. Angiogenesis Protocols: Second Edition, edited by Stewart Martin and Cliff Murray, 2008 466. Kidney Research: Experimental Protocols, edited by Tim D. Hewitson and Gavin J. Becker, 2008. 465. Mycobacteria, Second Edition, edited by Tanya Parish and Amanda Claire Brown, 2008 464. The Nucleus, Volume 2: Physical Properties and Imaging Methods, edited by Ronald Hancock, 2008 463. The Nucleus, Volume 1: Nuclei and Subnuclear Components, edited by Ronald Hancock, 2008 462. Lipid Signaling Protocols, edited by Banafshe Larijani, Rudiger Woscholski, and Colin A. Rosser, 2008 461. Molecular Embryology: Methods and Protocols, Second Edition, edited by Paul Sharpe and Ivor Mason, 2008 460. Essential Concepts in Toxicogenomics, edited by Donna L. Mendrick and William B. Mattes, 2008 459. Prion Protein Protocols, edited by Andrew F. Hill, 2008 458. Artificial Neural Networks: Methods and Applications, edited by David S. Livingstone, 2008 457. Membrane Trafficking, edited by Ales Vancura, 2008 456. Adipose Tissue Protocols, Second Edition, edited by Kaiping Yang, 2008 455. Osteoporosis, edited by Jennifer J. Westendorf, 2008 454. SARS- and Other Coronaviruses: Laboratory Protocols, edited by Dave Cavanagh, 2008 453. Bioinformatics, Volume 2: Structure, Function, and Applications, edited by Jonathan M. Keith, 2008 452. Bioinformatics, Volume 1: Data, Sequence Analysis, and Evolution, edited by Jonathan M. Keith, 2008 451. Plant Virology Protocols: From Viral Sequence to Protein Function, edited by Gary Foster, Elisabeth Johansen, Yiguo Hong, and Peter Nagy, 2008 450. Germline Stem Cells, edited by Steven X. Hou and Shree Ram Singh, 2008 449. Mesenchymal Stem Cells: Methods and Protocols, edited by Darwin J. Prockop, Douglas G. Phinney, and Bruce A. Brunnell, 2008 448. Pharmacogenomics in Drug Discovery and Development, edited by Qing Yan, 2008. 447. Alcohol: Methods and Protocols, edited by Laura E. Nagy, 2008 446. Post-translational Modifications of Proteins: Tools for Functional Proteomics, Second Edition, edited by Christoph Kannicht, 2008.

445. Autophagosome and Phagosome, edited by Vojo Deretic, 2008 444. Prenatal Diagnosis, edited by Sinhue Hahn and Laird G. Jackson, 2008. 443. Molecular Modeling of Proteins, edited by Andreas Kukol, 2008. 442. RNAi: Design and Application, edited by Sailen Barik, 2008. 441. Tissue Proteomics: Pathways, Biomarkers, and Drug Discovery, edited by Brian Liu, 2008 440. Exocytosis and Endocytosis, edited by Andrei I. Ivanov, 2008 439. Genomics Protocols, Second Edition, edited by Mike Starkey and Ramnanth Elaswarapu, 2008 438. Neural Stem Cells: Methods and Protocols, Second Edition, edited by Leslie P. Weiner, 2008 437. Drug Delivery Systems, edited by Kewal K. Jain, 2008 436. Avian Influenza Virus, edited by Erica Spackman, 2008 435. Chromosomal Mutagenesis, edited by Greg Davis and Kevin J. Kayser, 2008 434. Gene Therapy Protocols: Volume 2: Design and Characterization of Gene Transfer Vectors, edited by Joseph M. Le Doux, 2008 433. Gene Therapy Protocols: Volume 1: Production and In Vivo Applications of Gene Transfer Vectors, edited by Joseph M. Le Doux, 2008 432. Organelle Proteomics, edited by Delphine Pflieger and Jean Rossier, 2008 431. Bacterial Pathogenesis: Methods and Protocols, edited by Frank DeLeo and Michael Otto, 2008 430. Hematopoietic Stem Cell Protocols, edited by Kevin D. Bunting, 2008 429. Molecular Beacons: Signalling Nucleic Acid Probes, Methods and Protocols, edited by Andreas Marx and Oliver Seitz, 2008 428. Clinical Proteomics: Methods and Protocols, edited by Antonia Vlahou, 2008 427. Plant Embryogenesis, edited by Maria Fernanda Suarez and Peter Bozhkov, 2008 426. Structural Proteomics: High-Throughput Methods, edited by Bostjan Kobe, Mitchell Guss, and Thomas Huber, 2008 425. 2D PAGE: Sample Preparation and Fractionation, Volume 2, edited by Anton Posch, 2008 424. 2D PAGE: Sample Preparation and Fractionation, Volume 1, edited by Anton Posch, 2008 423. Electroporation Protocols: Preclinical and Clinical Gene Medicine, edited by Shulin Li, 2008 422. Phylogenomics, edited by William J. Murphy, 2008 421. Affinity Chromatography: Methods and Protocols, Second Edition, edited by Michael Zachariou, 2008 420. Drosophila: Methods and Protocols, edited by Christian Dahmann, 2008 419. Post-Transcriptional Gene Regulation, edited by Jeffrey Wilusz, 2008 418. Avidin–Biotin Interactions: Methods and Applications, edited by Robert J. McMahon, 2008

M E T H O D S I N M O L E C U L A R B I O L O G YT M

Essential Concepts in Toxicogenomics

Edited by

Donna L. Mendrick Gene Logic Inc, Gaithersburg, MD, USA

and

William B. Mattes The Critical Path Institute, Rockville, MD, USA

Editors Donna L. Mendrick Gene Logic Inc, Gaithersburg, MD, USA [email protected]

William B. Mattes The Critical Path Institute, Rockville, MD, USA [email protected]

Series Editor John M. Walker Professor Emeritus School of Life Sciences University of Hertfordshire Hatfield Hertfordshire AL10 9AB, UK [email protected]

ISBN: 978-1-58829-638-2

e-ISBN: 978-1-60327-048-9

Library of Congress Control Number: 2008921701 ©2008 Humana Press, a part of Springer Science+Business Media, LLC All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Humana Press, 999 Riverview Drive, Suite 208, Totowa, NJ 07512 USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. While the advice and information in this book are believed to be true and accurate at the date of going to press, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect to the material contained herein. Cover illustration: Chapter 2, Fig. 1. Printed on acid-free paper 987654321 springer.com

Preface

The field of toxicogenomics is moving rapidly, so it is impossible at the time of this writing to compile a classic methods textbook. Instead, we chose to identify experts in all aspects of this field and challenged them to write reviews, opinion pieces, and case studies. This book covers the main areas important to the study and use of toxicogenomics. Chapter 1 speaks to the convergence of classic approaches alongside toxicogenomics. Chapter 2 deals with the usefulness of toxicogenomics to identify the mechanism of toxicity. Chapter 3 calls attention to the issues that affect the quality of toxicogenomics experiments, as well as the implications of using microarrays as diagnostic devices. The need for appropriate statistical approaches to genomic data is discussed in Chapter 4, and Chapters 5 and 6 describe the use of genomic data to build toxicogenomic models and provide insights from the approaches of two companies. The important topic of storing the data generated in such experiments and the correct annotation that must accompany such data is considered in Chapter 7. The discussion in Chapter 8 speaks to the use of toxicogenomics to identify species similarities and differences. Chapters 9 and 10 deal with the use of genomics to identify biomarkers within the preclinical and clinical arenas. Biomarkers will only be useful if the community at large accepts them as meaningful. Consortia are important to drive this function, and Chapter 11 discusses current efforts in this area. Last but not least, Chapter 12 presents a perspective on the regulatory implications of toxicogenomic data and some of the hurdles that can be seen in its implication in GLP studies. Although this book tends to focus on pharmaceuticals, the issues facing toxicology are shared by the chemical manufacturers, the tobacco industry, and their regulators. We want to thank our contributors for their generous time and energy in providing their insights. Sadly, we must note the unexpected passing of one of our authors, Dr. Joseph Hackett of the FDA. Joe’s contribution serves as a testimony to his accomplishments in this field, and his insight will be missed in the years to come. Donna L. Mendrick William B. Mattes

v

Contents Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Color Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Toxicogenomics and Classic Toxicology: How to Improve Prediction and Mechanistic Understanding of Human Toxicity Donna L. Mendrick . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

v ix xi

1

2

Use of Traditional End Points and Gene Dysregulation to Understand Mechanisms of Toxicity: Toxicogenomics in Mechanistic Toxicology Wayne R. Buck, Jeffrey F. Waring, and Eric A. Blomme . . . . . . . . . . . . 23

3

Quality Control of Microarray Assays for Toxicogenomic and In Vitro Diagnostic Applications Karol L. Thompson and Joseph Hackett . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4

Role of Statistics in Toxicogenomics Michael Elashoff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5

Predictive Toxicogenomics in Preclinical Discovery Scott A. Barros and Rory B. Martin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

6 In Vivo Predictive Toxicogenomics Mark W. Porter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 7

Bioinformatics: Databasing and Gene Annotation Lyle D. Burgoon and Timothy R. Zacharewski . . . . . . . . . . . . . . . . . . . . . 145

8

Microarray Probe Mapping and Annotation in Cross-Species Comparative Toxicogenomics John N. Calley, William B. Mattes, and Timothy P. Ryan . . . . . . . . . . . 159

9

Toxicogenomics in Biomarker Discovery Marc F. DeCristofaro and Kellye K. Daniels. . . . . . . . . . . . . . . . . . . . . . . . 185

10

From Pharmacogenomics to Translational Biomarkers Donna L. Mendrick . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

11

Public Consortium Efforts in Toxicogenomics William B. Mattes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

vii

viii

Contents 12

Applications of Toxicogenomics to Nonclinical Drug Development: Regulatory Science Considerations Frank D. Sistare and Joseph J. DeGeorge . . . . . . . . . . . . . . . . . . . . . . . . . . 239

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

Contributors

Scott A. Barros • Toxicology, Archemix Corp., Cambridge, Massachusetts Eric A. Blomme • Department of Cellular and Molecular Toxicology, Abbott Laboratories, Abbott Park, Illinois Wayne R. Buck • Department of Cellular and Molecular Toxicology, Abbott Laboratories, Abbott Park, Illinois Lyle D. Burgoon • Department of Biochemistry & Molecular Biology, Michigan State University, East Lansing, Michigan John N. Calley • Department of Integrative Biology, Eli Lilly and Company, Greenfield, Indiana Kellye K. Daniels • Department of Toxicogenomics, Gene Logic Inc., Gaithersburg, Maryland Marc F. DeCristofaro • Biomarker Development, Novartis Pharmaceuticals Corporation, East Hanover, New Jersey Joseph J. DeGeorge • Laboratory Sciences and Investigative Toxicology, Merck & Co Inc, West Point, Pennsylvania Michael Elashoff • Department of BioStatistics, CardioDx, Palo Alto, California Joseph Hackett • Office of Device Evaluation, Center for Devices and Radiological Health, U.S. Food and Drug Administration, Rockville, Maryland Rory B. Martin • Drug Safety and Disposition, Millennium Pharmaceuticals, Cambridge, Massachusetts William B. Mattes • Department of Toxicology, The Critical Path Institute, Rockville, Maryland Donna L. Mendrick • Department of Toxicogenomics, Gene Logic Inc., Gaithersburg, Maryland Mark W. Porter • Department of Toxicogenomics, Gene Logic Inc., Gaithersburg, Maryland Timothy P. Ryan • Department of Integrative Biology, Eli Lilly and Company, Greenfield, Indiana Frank D. Sistare • Laboratory Sciences and Investigative Toxicology, Merck & Co Inc., West Point, Pennsylvania ix

x

Contributors

Karol L. Thompson • Division of Applied Pharmacology Research, Center for Drug Evaluation and Research, U.S. Food and Drug Administration, Silver Spring, Maryland Jeffrey F. Waring • Department of Cellular and Molecular Toxicology, Abbott Laboratories, Abbott Park, Illinois Timothy R. Zacharewski • Department of Biochemistry and Molecular Biology, Michigan State University, East Lansing, Michigan

Color Plates Color plates follow p. 112. Color Plate 1

Color Plate 2

Color Plate 3

Color Plate 4 Color Plate 5

Color Plate 6

Color Plate 7 Color Plate 8

Identification of genes regulated in the liver of rats after xenobiotic activation of the nuclear receptors PPAR, aromatic hydrocarbon receptor (AhR), or pregnane X receptor (PXR). (Chapter 2, Fig. 1; see legend and discussion on p. 26.) Hierarchical clustering of gene expression profiles of the testes of male Sprague-Dawley rats treated with a single dose of various testicular toxicants and sacrificed 24 h after treatment. (Chapter 2, Fig. 2; see legend and discussion on p. 35.) Heatmap of gene expression profiles from the liver of rats treated with Cpd-001 (arrow) and a wide variety of reference compounds including nonhepatotoxicants and hepatotoxicants. (Chapter 2, Fig. 4; see legend and discussion on p. 38.) Distributions for error estimators based on proteasome data. (Chapter 5, Fig. 2; see legend and discussion on p. 95.) Operating characteristics of the baseline in vitro classifier as a function of classification cutpoint. Replicate observations were treated independently. (Chapter 5, Fig. 5; see legend and discussion on p. 104.) Operating characteristics of the baseline in vivo classifier as a function of classification cutpoint. (Chapter 5, Fig. 6; see legend on p. 105 and discussion on p. 104.) Similarity tree for in vitro compounds. (Chapter 5, Fig. 7; see legend on p. 107 and discussion on p. 106.) Model scores for two doses of thioacetamide or vehicletreated samples at 6-, 24-, and 48-h exposures. (Chapter 6, Fig. 4; see legend and discussion on p. 138.)

xi

1 Toxicogenomics and Classic Toxicology: How to Improve Prediction and Mechanistic Understanding of Human Toxicity Donna L. Mendrick

Summary The field of toxicogenomics has been advancing during the past decade or so since its origin. Most pharmaceutical companies are using it in one or more ways to improve their productivity and supplement their classic toxicology studies. Acceptance of toxicogenomics will continue to grow as regulatory concerns are addressed, proof of concept studies are disseminated more fully, and internal case studies show value for the use of this new technology in concert with classic testing.

Key Words: hepatocytes; hepatotoxicity; idiosyncratic; phenotypic anchoring; toxicogenomics; toxicology.

1. Introduction The challenges facing the field of toxicology are growing as companies demand more productivity from their drug pipelines. The intent of this chapter is to identify the issues facing classic approaches to nonclinical toxicity testing, the cause of the deficiencies, and ways in which toxicogenomics can improve current in vitro and in vivo testing paradigms. The public at large continues to exert pressure on the pharmaceutical industry to develop new drugs yet are intolerant of safety issues and the high cost of drugs when they reach the market. This is setting up a “perfect storm” with a recognized decrease in productivity in this industry, continual increase in costs of developing new drugs, and rising attrition rates due to nonclinical and clinical safety failures (1–4). From: Methods in Molecular Biology, vol. 460: Essential Concepts in Toxicogenomics Edited by: D. L. Mendrick and W. B. Mattes © Humana Press, Totowa, NJ

1

2

Mendrick

% Terminated

2. Classic Toxicology Those in the field of drug and chemical development know of the multitude of compounds to which humans were never exposed either in the clinic or in the environment because of obvious toxicity seen in preclinical species. However, it is well-known that classic testing in animals is not infallible. A study done by a group within the Institutional Life Sciences Institute (ILSI) illustrates the problem (5). Twelve companies contributed data on 150 compounds that have shown toxicity in humans of a significant enough nature to warrant one of four actions: (1) termination, (2) limitation of dosage, (3) need to monitor drug level, or (4) restriction of target population. The group compared the human toxicities with the results of the classic toxicity employed for each drug. They found that only ∼70% of these toxicities could be predicted in classic animal testing even when multiple species, primarily the rat (rodent) and dog (nonrodent), were employed. The dog was better than the rat in predicting human toxicity (63% vs. 43%, respectively), with the success rate varying depending on the human target organ. However, escalating concerns regarding the use of animals in medical research, the amounts of compound required for such large animals, and the cost of such studies prevents this species from being used as the first species or in sufficient numbers to detect subtle toxicities. The exact failure rate due to toxicity and the time of its detection continues to be the subject of study because only by understanding the problem can one begin to propose solutions. Authors tend to report somewhat different findings. The drugs terminated because of human toxicities evaluated in the ILSI study (5) failed most often during Phase II (Fig. 1). Suter et al. at Roche

45 40 35 30 25 20 15 10 5 0

43 39

10

Phase I

Phase II

Phase III

Fig. 1. Data illustrating the termination rate of compounds due to human toxicity during clinical trials. (Data adapted from Olson, H., Betton, G., Robinson, D., Thomas, K., Monro, A., Kolaja, G., et al. 2000. Concordance of the toxicity of pharmaceuticals in humans and in animals. Regul. Toxicol. Pharmacol. 32, 56–67.)

% Terminated

Toxicogenomics and Classic Toxicology

3

50 45 40 35 30 25 20 15 10 5 0

al

nic

li rec

P

eI

as

Ph

se

a Ph

II

se

a Ph

III R

n

tio

tra

is eg

Clinical Safety Animal Toxicology

Fig. 2. Failure rate due to animal toxicity and human safety during the development pipeline. (Data adapted from Suter, L., Babiss, L.E., and Wheeldon, E.B. 2004. Toxicogenomics in predictive toxicology in drug development. Chem. Biol. 11, 161–171.)

(6) examined the failure rate of compounds from preclinical to registration and divided safety failures into animal toxicity and human toxicities. Their work found the highest failure rates due to human safety in Phase I and registration (Fig. 2). Note that both studies found a high rate of failure in Phase II or beyond, a costly scenario. Dimasi and colleagues have examined financial models of drug development and have estimated the savings of terminating unsafe compounds earlier within the clinical trial paradigm (7). For example, if 25% of the drugs that will fail in Phase II were discontinued in Phase I, the clinical cost savings alone per approved drug would be $13 million to $38 million dollars. Obviously, the cost savings will be greater if one could prevent such a drug from even entering clinical trials by improving preclinical detection and/or by failing it earlier within the clinical testing phase (e.g., Phase I vs. Phase III). The Food and Drug Administration’s (FDA) Critical Path Initiative (www.fda.gov/oc/initiatives/criticalpath/) quotes one company executive as saying clinical failures due to hepatotoxicity had cost the company more than $200 million per year over the past decade. Clearly, there are many financial incentives to address the issue of safety. 3. Toxicogenomics Many in the field have written excellent opinion pieces and reviews on the use of toxicogenomics in drug discovery and development and in the chemical/agrochemical sectors. Toxicogenomics is used in three areas: predictive applications for compound prioritization, mechanistic analyses for compounds with observed toxicity, and biomarker identification for future

4

Mendrick

screens or to develop biomarkers useful in preclinical and/or clinical studies. Though impractical to list all of the relevant publications, a few excellent articles on toxicogenomics are provided (2,3,6,8–15). 3.1. Study Designs As with all scientific endeavors, to answer the questions being posed it is important to have an optimal study design. Genomics tends to be somewhat expensive, so understandably some try to downsize the experimental setting. Unfortunately, that may prevent hypothesis generation or evaluation of a preexisting theory. As an example, if one is trying to form a hypothesis as to the mechanism of injury induced by a compound, sampling tissue only at the time of such damage may prevent evaluation of the underlying events that started the pathologic processes. Similarly, sampling only one time point will inhibit the fullest evaluation of the dynamic processes of injury and repair. In classic toxicology testing, one would not claim with certainty that a compound is not hepatotoxic if one saw no elevation of serum alanine aminotransferase (ALT) or histologic change in the liver in a snapshot incorporating only one time point and dose level. Likewise, one does not pool blood from all animals and perform clinical pathology on such. Unfortunately, some have approached toxicogenomics in this manner (using restrictive study designs and pooled RNA samples) and then felt betrayed by the lack of information. This does not mean to suggest that all toxicogenomic studies must be all-encompassing as long as the investigator understands beforehand the limits of his or her chosen design. One approach might be to collect samples from multiple time points and doses and triage the gene expression profiling to determine the most important study groups within that experimental setting. Establishment of the appropriate dose is important as well. Classic toxicology endeavors use dose escalation until one sees a phenotypic adverse event such as changes in classic clinical pathology, histology, body weight, and so forth. An anchoring of the dose used for toxicogenomic studies also must be employed and contextual effects of such doses understood. Doses that severely affect body weight likely induce great stress upon the animal, and this must be taken into account if this phenotypic anchor is followed. A recent paper by Shioda et al. studied effects of xenoestrogens in cell culture and explored the relationship of doses to transcriptional profiles (16). This work suggests that doses chosen for equivalent cellular responses highlight the differences between compounds while those selected based on the compounds’ action on a particular gene reveal mostly similarities between the compounds. Additional work remains to be done to determine if this conclusion can be extrapolated to other compound types and in vivo environments, but, at a minimum, this report reinforces the

Toxicogenomics and Classic Toxicology

5

need to understand the chosen study design and fully explain criteria used for dose selection. 3.2. Genomic Approaches Can Clarify Basic Husbandry Issues Genomics enables detection of toxicity parameters as well as differences in animal husbandry. In many cases, the study design may call for food restriction or animals may be accidentally deprived of food. Genomic analysis can detect such events as shown in Fig. 3 and Fig. 4. Studies in Gene Logic’s (Gaithersburg, MD) ToxExpress® database were used for the analysis. In Fig. 3, the data from the probe sets (∼8800) on the Affymetrix Rat Genome U34A GeneChip® microarray (Santa Clara, CA) were subjected to a principal components analysis (PCA). Such a test illustrates underlying differences in the data in a multidimensional picture. For ease in viewing, two-dimensional graphical representations are provided. In Fig. 3, the data from all probe sets were used, and, even with the accompanying noise when so many parameters are measured, one can see differentiation of the groups particularly if one combines the x and y axes, accounting for 39% of the gene expression variability. Treatment Fasted Fed

PCA Mapping (39.3%) 41 31 22

PC #2 14.3%

13 4 –5 –14 –23 –32 –41 –51 –80

–64

–49

–34

–19

–5 10 PC #1 25%

25

40

55

70

Fig. 3. A PCA using all genes on the Rat Genome U34A microarray illustrates the differentiation on a genomic basis between rats fasted for 24 h versus those rats that had food ad libitum. Use of all genes on the array is accompanied by noise and yet one obtains reasonable separation using the x axis and better discrimination if one employs both x and y axes. Such findings illustrate the ability of gene expression to provide insight into animal husbandry.

6

Mendrick Effect of Fasting Fasted Fed

PCA Mapping (81.5%) 4 2.6 1.2

PC #2 5.6%

–0.2 –1.6 –3 –4.4 –5.8 –7.2 –8.6 –10 –16

–12.9

–9.8

–6.7

–3.6 –0.5 2.6 PC #1 75.9%

5.7

8.8

11.9

15

Fig. 4. An analysis filter was applied to identify differentially regulated genes. The filter has a cutoff as follows: fold-change ≥1.8 with t-test p-value 40. This resulted in 281 genes identified as dysregulated between fed and fasted rats. Almost all of the variation is captured in PC #1 (75.9%) and the groups are more clearly separated than seen when all genes were used as shown in Fig. 3.

When genes that were differentially regulated among the fed and fasted animals were chosen, the gene list was reduced from >5000 to 281. The results in Fig. 4 demonstrate a complete discrimination of these rats with the x axis accounting for 76% of the variability. Genomics can be used to discriminate strain and gender as well. In the former case, female rats of Sprague-Dawley (SD) or Wistar origin were compared. Although evaluation using all genes discriminated these strains (data not shown), selection of differentially regulated genes resulted in a clearer separation as shown in Fig. 5. Because both strains are albino, one could envision using a genomics approach should there be a potential mix-up of strains in the animal room. What is likely less surprising is the ability to categorize gender based on gene expression findings. Although it is usually easy to identify the gender of rats by physical examination alone, one could envision the use of a genomic approach to study the feminization of male rats under drug treatment or vice versa. As shown in Fig. 6, a PCA employing all genes discriminates between genders although the first two axes capture only ∼23% of the variation suggesting there

Toxicogenomics and Classic Toxicology

7 Effect of Strain SD Wistar

PCA Mapping (69.9%) 4.1 3.16 2.22

PC #2 6.72%

1.28 0.34 –0.6 –1.54 –2.48 –3.42 –4.36 –5.3 –9

–7.3

–5.6

–3.9

–2.2 –0.5 1.2 PC #1 63.2%

2.9

4.6

6.3

8

Fig. 5. A PCA was performed with 83 genes that were differentially expressed between normal female Sprague-Dawley (SD) or Wistar rats. Such strains are clearly discriminated using a genomic approach. Gender Female Male

PCA Mapping (24.3%) 44 34 24

PC #2 7.44%

14 4 –5 –14 –24 –34 –44 –54 –90

–72

–55

–38

–21 –5 12 PC #1 16.9%

29

46

63

80

Fig. 6. A PCA was performed with all genes from male and female rat livers. There is an overall separation particularly if one combines the variability shown along the x and y axes.

8

Mendrick Effect of Gender Female Male

PCA Mapping (68.9%) 6 4.9 3.8

PC #2 5.3%

2.7 1.6 0.5 –0.6 –1.7 –2.8 –3.9 –5 –14

–11.3

–8.6

–5.9

–3.2 –0.5 2.2 PC #1 63.6%

4.9

7.6

10.3

13

Fig. 7. A PCA was performed with 175 genes that were differentially expressed between the genders. In this case, the majority of difference was captured in the first axis resulting in a very clear separation of the genders.

are many subtle effects, a point not likely to be argued by many humans. Using genes differentially regulated, however, found that 175 genes can account for more than 63% of variation in the first axis alone as shown in Fig. 7. In the three cases described above (food deprivation, strain, and gender), differentially regulated genes were identified that enabled clear categorization between the groups. However, the reverse is also true. Removal of genes that identify such differences from the analysis can enable other differences to become more apparent. Such has been accomplished with the predictive models built as part of our ToxExpress program specifically to avoid such confounding variables and enable such models to work well when female or male rats are used of various strains and independent of fasting. 3.3. Toxicogenomics Can Augment Understanding of Classic Toxicology Findings Toxicogenomics can add value to classic testing when one animal appears to have an aberrant finding. It is recognized that preclinical testing of drugs is a complicated process and mistakes can happen. If a rat did not exhibit any signs of toxicology unlike its cohorts, it would be informative to differentiate dosing errors from true differences in toxicity responses. One could monitor

Toxicogenomics and Classic Toxicology

9

the blood to determine if the drug was detectable, but that would depend on its clearance. Alternatively, one could explore gene expression as a subtle detection method. Figure 8 illustrates how similarity in gene expression can identify the treatment given. If nine rats received carbon tetrachloride and the tenth rat was left untreated, the gene expression of the latter would appear similar to that of untreated rats as shown in the upper left corner in this mock illustration. As seen with these examples, molecular approaches can be more sensitive in terms of discriminating animal husbandry, strain differences, and so forth, than parameters monitored in classic toxicity testing. Microarrays monitor tens of thousands of events (expression levels of genes and Expressed Sequence Tags (ESTs)), whereas classic toxicity testing has been estimated to measure approximately 100 parameters (17). From a strictly statistical approach, one can envision that more knowledge would lead to improved decision making although the challenge of removing the noise when monitoring so many events is real. Arrays provide information on the changes in individual genes, and from this one can then hypothesize on the effects on biological pathways, and

CCI4 24 hrs CornOil 24 hrs Untreated

50 PCA Mapping (36.5%) 38 26

PC #2 12.5%

14 2 –10 –21 –33 –45 –57 –70 –55

–41

–27

–13 0 14 PC #1 24.1%

28

42

56

70

Fig. 8. A PCA illustrates the greatest source of difference at the gene expression level between the rats. The first component (PC #1) accounts for 24% of the variability, and rats treated with carbon tetrachloride are clearly delineated from those receiving vehicle or left untreated with the exception of the one rat in the upper left. The second component, PC #2, accounts for 12.5% of the variability and discriminates even untreated from vehicle-treated rats further illustrating the sensitivity of this approach.

10

Mendrick

so forth, a level of understanding not provided in classic approaches. One could use the analogy of the sensitivity of electron versus light microscopy. Disease processes such as minimal change nephropathy induce severe functional alterations (proteinuria) in the human kidney yet are extremely difficult to visualize with light microscopy. However, the morphologic changes associated with this proteinuria are clearly seen in such diseased kidneys if one uses electron microscopy, a more sensitive technique. At the ultrastructural level, a flattening of visceral epithelial cells upon the glomerular basement membrane is clearly visible, and such a process is known to be associated with proteinuria in preclinical species and in humans. Several recent papers highlight the improved sensitivity of a genomics approach with classic examples of compounds that produce adverse events in rats. Heinloth and her colleagues at the National Institute of Environmental Health and Safety (NIEHS) have examined the dose response relationship of rat liver to acetaminophen (18). They studied multiple doses in the rat using two (they called these “subtoxic”) doses that failed to cause changes monitored with classic approaches. However, when ultrastructural methods were employed, some toxicity-associated mitochondrial changes were observed at one of these “subtoxic” doses. Evaluation of gene expression changes found, as a consequence of acetaminophen toxicity, a loss in cellular energy even at such “subtoxic” doses. As the dose was increased, so was the magnitude of changes observed, and this was accompanied by alterations in associated genes. They concluded that gene expression profiling may provide more sensitivity for detecting adverse effects in the absence of the occurrence of overt toxicity. Another recent paper explored a time savings approach. Nie and his colleagues at Johnson and Johnson Pharmaceutical Research and Development examined changes in gene expression in the rat liver at 24 h after exposure to nongenotoxic carcinogens and control compounds (19). They identified six genes that predicted 89% of nongenotoxic carcinogens after 1 day of exposure instead of the 2 years of chronic dosing normally required for such cellular changes to be clear. They also mined the gene expression results to find biologically relevant genes for a more mechanistic approach to understanding nongenotoxic carcinogenicity. Together the work by Heinloth et al. and Nie et al. highlight the sensitivity of a genomics approach to detect compounds that will elicit classic adverse events in rats either at higher doses or exposure times. Toxicogenomics can be employed to identify species-specific changes providing support for safety claims in humans. Peroxisome proliferator activated receptor alpha (PPAR-) agonists have been widely studied as they have been found to be clinically useful as hypolipidemic drugs yet induce hepatic tumors in rats. It is now appreciated that such an effect has little safety

Toxicogenomics and Classic Toxicology

11

risk to humans and that genes can be identified in rat liver and hepatocytes that enable discrimination of this mechanism. As suggested by Peter Lord and his colleagues at Johnson and Johnson, this could be incorporated into the safety evaluation of novel compounds by employing expression profiling of rat liver or hepatocytes exposed to the new drug. Similarities between this novel compound and known PPAR- agonists could provide a safety claim for the former’s species-specific effects (15). In some cases, differences in species responses to drugs may rely on dissimilarities in drug metabolism enzymes. A study was performed using data in Gene Logic’s ToxExpress database. Expression levels of genes involved in glutathione metabolism were compared among normal tissues and genders of the rat, mouse, and canine. As expected, expression levels varied among genes and tissues. Little differences were seen among male and female animals but large effects were seen among species. If differential gene expression does translate to enzymatic activities differences between species, it would be expected to impact drug responses. Such information could provide guidance into species selection for toxicity testing (20). Chapters later in this book discuss species differences in more detail. 3.4. Use of Toxicogenomics to Detect Idiosyncratic Compounds A well-accepted definition of the term idiosyncratic, particularly on a compound by compound basis, is hard to find as individuals use different criteria. Some depend on incidence alone, some on the inflammatory response induced, some on a lack of dose response relationship, and so forth. However, universal to most users’ definitions is the lack of classic changes observed in preclinical species. To complete the analogy started above in discussing the sensitivity differences between light and electron microscopy, one could suppose that idiosyncratic compounds elicit some effects at the gene and protein level in rats but these events do not lead to changes severe enough to induce classic phenotypic changes in this commonly used testing species. The failure to cause overt injury in the rat might be due to (1) the inability of this species to metabolize the drug into the toxic metabolite responsible for human injury, (2) quantitative differences whereby rats generate the toxic metabolite but at lower levels than seen in humans, (3) a superior ability of the rat to detoxify toxins, and (4) the failure of rats to respond in an immune-mediated manner to adduct formation of intrinsic cell surface proteins. Many if not all compounds today are commonly screened for metabolism differences between species and their development discontinued if such metabolic specific responses are seen. Even so, the postmarketing occurrence of drug hepatotoxicity is a leading cause of regulatory actions and further erodes the pharmaceutical pipeline. The field of

12

Mendrick

drug development needs new methods to identify such drugs earlier in discovery and development that can inform compound selection, position drugs to areas of unmet clinical need, and influence preclinical and clinical trial designs in a search for earlier biomarkers of liver injury induced by such a compound. Several groups including those from Millennium (Chapter 5 and Ref. 21), AstraZeneca (22), and Gene Logic (Chapter 6) have reported success in (a) training predictive models using all genes and ESTs being monitored on the array platform and (b) classifying the compounds based on their potential to induce toxicity in humans including idiosyncratic compounds. The result is sets of genes and ESTs that may not be easily translated into mechanistic understanding on their own yet show robust predictive ability. Such an approach has enabled the prediction of idiosyncratic hepatotoxicants from rats and rat hepatocytes exposed to such agents. Figure 9 illustrates the predictive ability of models built at Gene Logic to identify idiosyncratic compounds as potential human hepatotoxicants using gene expression data obtained from in vivo and in vitro exposure to prototypical compounds. The classification of individual compounds as idiosyncratic tends to be controversial, so to remove any internal bias, the statistics were compiled using marketed compounds classified as idiosyncratic by Kaplowitz (23). Kaplowitz subclassifies idiosyncratic compounds as allergic or nonallergic. As can be seen in Fig. 9, toxicogenomic predictions performed from in vivo rat exposure are equally accurate regardless of whether or not the idiosyncratic compound induces an allergic responses in the human liver, using Kaplowitz’s classification. In contrast, the model built from rat hepatocyte data does somewhat better with compounds that

% Predicted as Hepatotoxic in Humans

100

92% 92%

92% 92%

88%

92% 78% 78%

80

71% 71%

60

In Vivo Exposure in Rats

40

In Vitro Exposure to Primary Rat Hepatocytes

20 0 Allergic

Non-allergic

Allergic & Non-allergic

Fig. 9. Compounds described as idiosyncratic by Kaplowitz (23) were studied. He discriminates between drugs that induce allergic-type reactions from those that do not, and the results of the predictive toxicogenomic modeling are shown here.

Toxicogenomics and Classic Toxicology

13

cause nonallergic responses in humans. It should be remembered that idiosyncratic compounds do not elicit classic phenotypic changes in rats and other nonclinical species, yet the majority are detected using a predictive toxicogenomics approach in rat liver and primary rat hepatocytes. How is it possible for genomics to predict something in the rat that is not seen at a classic phenotypic level?

3.4.1. Case Studies of Idiosyncratic Hepatotoxicants For most of the idiosyncratic drugs on the market today, little is known about their ability to induce hepatotoxicity in humans while avoiding classic detection in preclinical species. Felbamate is an exception, and it has been researched extensively. It has been reported that its metabolism is quantitatively different between species and this is believed to be responsible for the failure of classic testing to detect its potential to induce human hepatotoxicity. Researchers have reported that felbamate is metabolized to a reactive aldehyde (atropaldehyde) and that rats produce far less of this toxin than do humans (24). However, because rats do produce some level of this toxic metabolite, it raises the possibility that a technology more sensitive than classic testing (e.g., histopathology) may detect subtle signs of toxicity in rat. To test this hypothesis, an in vivo study done at Gene Logic employed 45 rats in groups of five per dose and time point. Animals were dosed daily by oral gavage with vehicle, low-dose or highdose felbamate. Groups were sacrificed at 6 h and 24 h after one treatment and 14 days after daily dosing. Histology of the livers was normal. One of the highdose–treated rats sacrificed after 14 days of exposure exhibited an ALT level above the normal reference range, and the entire group had a minor but statistically significant elevation above the control group. The overall conclusion of the board-certified veterinary pathologist was that a test article effect was unclear. Because this drug failed to demonstrate hepatotoxicity in the classic testing employed for registration, this result was not surprising. However, it does raise the issue of whether individual rats are responding in a classic sense to idiosyncratic compounds, and these results are not being interpreted correctly as suggested by retrospective analysis of some idiosyncratic compounds by Alden and colleagues at Millennium Pharmaceuticals (25). To explore the potential effects of felbamate on gene expression, livers were collected from all rats in this study, processed using Affymetrix GeneChip microarrays, and the data passed through the predictive toxicogenomics models built at Gene Logic. The results of the models, illustrated in Fig. 10, were correct in identifying felbamate as a potential human hepatotoxicant, a potential hepatitis-inducing agent in humans (26,27), and an agent that will cause liver enlargement in rats

14

Mendrick Toxicogenomic Models Questions Addressed General Model

Pathology/Mechanistic Models

Compound Assessment

Model Results

Might it induce human hepatotoxicity?

YES

What type of injury may it induce in humans and/or rats

Hepatitis (human); liver enlargement (rat)

What well-known compounds does it resemble?

E.g., nefazodone, an idiosyncratic compound

Fig. 10. Predictive toxicogenomic models were built at Gene Logic and validated internally and with customer-provided blinded data. The results of the genomic data obtained from felbamate-treated rat liver are shown here.

(28). The gene expression data from such treated rats showed some similarity to other idiosyncratic drugs such as nefazodone. Another interesting idiosyncratic drug is nefazodone. This oral antidepressant was approved in 1994 in the United States, but postmarket cases of idiosyncratic adverse reactions were reported that resulted in 20 deaths. A black box warning was added to the label in 2001 and the drug withdrawn in 2004 (23,29–34). The only hepatic effect reported in rats was a dose-related (50, 150, 300 mg kg−1 day−1 ) increase in liver weight reported in a 13-week rat study (SBA NDA 20-152). Several mechanisms for its toxicity have been proposed. The first hypothesis regarding its toxicity is a generation of reactive metabolites via bioactivation, which are capable of covalently modifying CYP3A4, a drugmetabolizing enzyme it is metabolized by and inhibits (35–37). The second theory is that nefazodone compromises biliary elimination, resulting in an increase in drug and bile acids retained in the liver over time (34,38). These authors have found a transient increase in serum bile acids in rats suggesting that rats are responding to this drug inappropriately but can recover from its toxicity prior to developing classic phenotypic changes. To investigate whether gene expression changes might be able to detect nefazodone as a potential human hepatotoxicant in the absence of classic changes in the rat, a study was performed using doses of 0, 50, and 500 mg kg−1 day−1 given by oral gavage. Note that the high dose is only 67% higher than what the manufacturer used in its 13-week rat study in which liver enlargement was reported. Groups of rats were sacrificed 6 h and 24 h

Toxicogenomics and Classic Toxicology

15

after the first dose and 7 days after daily dosing. Clinical pathology and liver histopathology revealed no clear evidence of liver toxicity as was expected of this idiosyncratic compound. The results from toxicogenomic modeling were similar to that of felbamate, namely that it is a potential human hepatotoxicant that can cause hepatitis in humans and liver enlargement in rats and has similarity to idiosyncratic compounds such as felbamate. Kellye Daniels, Director of Toxicogenomics at Gene Logic, examined the gene expression data in more detail and found evidence for a transitory effect on Phase I and II drug-metabolizing enzymes, oxidative stress response genes, protein-repair associated genes, and a solute carrier. Taken together, this suggests the cells are metabolizing the conjugates for removal, proteins are damaged, and the cell is trying to remove them alongside a compensatory induction of a solute carrier to export them. These changes may reflect the reason rats do not develop obvious hepatotoxicity. It is hypothesized that idiosyncratic compounds have a similar phenotypic pattern at the gene level to compounds known to induce hepatotoxicity in rats and that enables predictive models to detect the former as well as the latter. The effect of compounds on the expression of superoxide dismutase 2 (Sod2), shown in Fig. 11, will serve as an example. This gene encodes a protein that catalyzes the breakdown of superoxide into hydrogen peroxide and water in the mitochondria and is therefore a central regulator of reactive oxygen species (ROS) levels (39). The expression of this gene is upregulated by compounds such as acetaminophen and lipopolysaccharide, known to induce

Cross-Compound Comparison of Sod2 Expression Rosiglitazone Diclofenac

*

Acetaminophen

*

Lipopolysaccharide

*

Nefazodone

* –2

–1

0

1

2

3

4

Fold Change (24 hr exposure)

5

6

*p < 0.05

Fig. 11. The expression level of superoxide dismutase 2 (Sod2) in toxicant-treated liver tissue compared with vehicle controls is illustrated. Both the fold change and statistical results are shown.

16

Mendrick

hepatotoxicity in multiple species yet acting by differing mechanisms. Two idiosyncratic compounds, diclofenac and nefazodone, also elevate the level of this gene, whereas a drug not associated with hepatotoxicity, rosiglitazone, has no significant effect. This illustrates the commonality of gene expression among many compounds that cause human hepatotoxicity whether or not they induce similar toxicity in rats, the species examined here. 4. Use of Toxicogenomics in an In Vitro Approach Because earlier identification of a compound destined to fail in the marketplace would save significant resources, it is important to examine what can be done in terms of both cell culture (in vitro) and animal (in vivo) testing. Although cells in a culture environment do not maintain their in situ morphology, their homeostatic function and potentially the full complement of drug-metabolizing genes, the ease of use, and relative savings in terms of compound and time continues to induce researchers to exploit in vitro systems. Companies tend to focus on cell culture approaches to rank compounds on the basis of potential toxicity during lead optimization but only if sufficient throughput and cost constraints can be met (3,40). Because the in vitro environment does not represent all of the complex processes at play upon in vivo exposure, during lead prioritization some prefer to focus on short-term in vivo approaches instead (41). How can toxicogenomics add value to classic in vitro approaches (e.g., cytotoxicity) and in vivo studies? First, it is helpful to identify what is needed in these stages. Donna Dambach and her colleagues at Bristol-Myers Squibb were working with a tiered testing strategy using immortalized human hepatocyte cell lines and primary rat and human hepatocytes (40). They report good success in using their five human hepatocyte cell lines in a cytotoxicity assay wherein compounds with an IC50 value of 2